search

yl4579 / StyleTTS2

StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models
MIT License
4.98k stars 422 forks source link

OOD data for LibriTTS-460 training? #187

Closed kmn1024 closed 8 months ago

kmn1024 commented 10 months ago

Related to https://github.com/yl4579/StyleTTS2/issues/95, if the existing OOD_data is based on LibriTTS and used during LJSpeech training, what OOD_data did you use for LibriTTS training?

yl4579 commented 8 months ago

It was the same as LibriTTS-460, so the OOD data is the same as itself, although the speakers are different. This is because the dataset is already large enough so we don't need an extra OOD dataset (though it helps if you can get more).